人工神经网络-近红外光谱法在线测定白砂糖粒度

On-line determination of white granulated sugar granularity by artificial neural network-near infrared spectroscopy

采用国产漫反射近红外在线分析系统,对广西某糖厂的白砂糖粒度进行了在线测定研究。采用SupNIR-4000型近红外在线分析仪,直接对成品传送带上的白砂糖进行扫描获得光谱数据,测定时间仅需90s。光谱预处理方法"标准化+Savitzky-Golay一阶求导+正交信号校正"效果最佳,人工神经网络(ANN)算法建立模型最佳。结果显示,ANN模型的校正均方根误差(RMSEC)为2.9718,预测均方根误差(RMSEP)为3.5244,粒度预测偏差满足糖厂要求的±5。此法在制糖行业具有广泛的应用前景。

On-line determining white granulated sugar granularity of a sugar factory in Guangxi was carried out using a domestic online near infrared spectroscopic analysis system （SupNIR-4000 NIR analyzer）. The spectral data were obtained by scanning the granulated sugar on the belt conveyer with NIR analyzer, the time measured only 90 s. The spectral preprocessing method ＂Standardization ＋ Savitzky-Golay 1st derivative ＋ orthogonal signal correction （OSC）＂ was the best and artificial neural network （ANN） modeling was the best. The results showed that ANN model root mean square error of calibration（RMSEC） was 2.9718, the root mean square error of prediction（RMSEP） was 3.5244. the prediction errors was ±5, to meet the sugar factory’s requirement. This method has a wide application prospect in the sugar industry.